Optimize for Voice

What you can ship

Use this pattern whenever speech is the product: voicemail, support call archives, IVR prompts, narration, and voice memos. Those sources are usually mono with narrow dynamics — keeping a multi-megabyte WAV (or a music-oriented HQ AAC) forever burns storage and bandwidth for almost no perceptual gain. A dedicated voice variant gives every upload a predictable, speech-tuned file you can hand to players, download endpoints, or cold storage.

That is roughly a 20×–40× shrink versus the source WAV before you delete the original. At library scale — thousands of messages — you move from “keep everything forever” to “keep what listeners actually play.”

1. Upload any quality voice recording

Skip client-side crush. Phone apps, Zoom exports, cheap recorders, and studio WAVs can all land in the same collection — AudioDN probes the file and builds variants from the best available source. Prefer the highest quality you have at upload time (WAV / FLAC / AIFF when possible) so the Opus encode starts from a clean signal.

This walkthrough uses a generated answering-machine sample — a classic beep followed by an overbearing mother asking, with maximum comic timing, when the grandchildren are happening. Download the source WAV if you want to run the upload yourself: answering-machine-message.wav.

Create an upload session

# COLLECTION_ID: uuid of the collection to upload into
curl -X POST "https://api.audiodelivery.net/v1/upload_session" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "collection_id": "COLLECTION_ID",
  "expires_in": 7200
}'

// COLLECTION_ID: uuid of the collection to upload into

const payload = {
  "collection_id": "COLLECTION_ID",
  "expires_in": 7200
};

const response = await fetch('https://api.audiodelivery.net/v1/upload_session', {
  method: 'POST',
  headers: {
    'Authorization': 'Bearer YOUR_API_KEY',
    'Content-Type': 'application/json',
  },
  body: JSON.stringify(payload)
});

const data = await response.json();

# COLLECTION_ID: uuid of the collection to upload into

import requests

headers = {
    "Authorization": "Bearer YOUR_API_KEY",
    "Content-Type": "application/json",
}

payload = {
    "collection_id": "COLLECTION_ID",
    "expires_in": 7200
}

response = requests.post(
    "https://api.audiodelivery.net/v1/upload_session",
    headers=headers,
    json=payload,
)
data = response.json()

Create a track in the session

# SESSION_ID: upload_session_id from the previous call
curl -X POST "https://api.audiodelivery.net/v1/upload/SESSION_ID/track" \
  -H "Content-Type: application/json" \
  -d '{
  "file_name": "answering-machine-message.wav"
}'

// SESSION_ID: upload_session_id from the previous call

const payload = {
  "file_name": "answering-machine-message.wav"
};

const response = await fetch('https://api.audiodelivery.net/v1/upload/SESSION_ID/track', {
  method: 'POST',
  headers: {
    'Content-Type': 'application/json',
  },
  body: JSON.stringify(payload)
});

const data = await response.json();

# SESSION_ID: upload_session_id from the previous call

import requests

headers = {
    "Content-Type": "application/json",
}

payload = {
    "file_name": "answering-machine-message.wav"
}

response = requests.post(
    "https://api.audiodelivery.net/v1/upload/SESSION_ID/track",
    headers=headers,
    json=payload,
)
data = response.json()

Upload the file bytes

# PUT the recording bytes to track_upload.upload_url.
# Any supported audio works — start with the highest quality you have (WAV shown here).
curl -X PUT "$UPLOAD_URL" \
  --data-binary @answering-machine-message.wav \
  -H "Content-Type: audio/wav"

Poll GET /v1/track/:track_id until track_status_id is ready. Organization-wide variants finish during this pass — including an Original Upload file if that recipe is enabled. Upload session details: Upload Sessions API.

2. Transcode to a voice Opus variant

Create a transcode variant named voice with codec: “opus”. For speech, start at 32 kbps mono — dramatically smaller than music presets, still intelligible on earbuds and handsets. Bump to 64 kbps when you need more fidelity for noisy sources. Strip tags and cover art; they only waste bytes on voicemail.

Configure the recipe once in the dashboard so every upload gets a voice file automatically, or append it to a ready track with the Variants API (below). Only lossy codecs are allowed on this endpoint — Opus is exactly what you want here.

Recommended: 32 kbps mono Opus

# TRACK_ID: uuid of the ready track
curl -X POST "https://api.audiodelivery.net/v1/track/TRACK_ID/variant" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "variant_type_id": "transcode",
  "index": "voice",
  "codec": "opus",
  "bitrate": 32,
  "is_stereo": false,
  "is_strip_tags": true,
  "is_strip_cover": true
}'

// TRACK_ID: uuid of the ready track

const payload = {
  "variant_type_id": "transcode",
  "index": "voice",
  "codec": "opus",
  "bitrate": 32,
  "is_stereo": false,
  "is_strip_tags": true,
  "is_strip_cover": true
};

const response = await fetch('https://api.audiodelivery.net/v1/track/TRACK_ID/variant', {
  method: 'POST',
  headers: {
    'Authorization': 'Bearer YOUR_API_KEY',
    'Content-Type': 'application/json',
  },
  body: JSON.stringify(payload)
});

const data = await response.json();

# TRACK_ID: uuid of the ready track

import requests

headers = {
    "Authorization": "Bearer YOUR_API_KEY",
    "Content-Type": "application/json",
}

payload = {
    "variant_type_id": "transcode",
    "index": "voice",
    "codec": "opus",
    "bitrate": 32,
    "is_stereo": False,
    "is_strip_tags": True,
    "is_strip_cover": True
}

response = requests.post(
    "https://api.audiodelivery.net/v1/track/TRACK_ID/variant",
    headers=headers,
    json=payload,
)
data = response.json()

Optional: 64 kbps if you need more headroom

# TRACK_ID: uuid of the ready track
curl -X POST "https://api.audiodelivery.net/v1/track/TRACK_ID/variant" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "variant_type_id": "transcode",
  "index": "voice",
  "codec": "opus",
  "bitrate": 64,
  "is_stereo": false,
  "is_strip_tags": true,
  "is_strip_cover": true
}'

// TRACK_ID: uuid of the ready track

const payload = {
  "variant_type_id": "transcode",
  "index": "voice",
  "codec": "opus",
  "bitrate": 64,
  "is_stereo": false,
  "is_strip_tags": true,
  "is_strip_cover": true
};

const response = await fetch('https://api.audiodelivery.net/v1/track/TRACK_ID/variant', {
  method: 'POST',
  headers: {
    'Authorization': 'Bearer YOUR_API_KEY',
    'Content-Type': 'application/json',
  },
  body: JSON.stringify(payload)
});

const data = await response.json();

# TRACK_ID: uuid of the ready track

import requests

headers = {
    "Authorization": "Bearer YOUR_API_KEY",
    "Content-Type": "application/json",
}

payload = {
    "variant_type_id": "transcode",
    "index": "voice",
    "codec": "opus",
    "bitrate": 64,
    "is_stereo": False,
    "is_strip_tags": True,
    "is_strip_cover": True
}

response = requests.post(
    "https://api.audiodelivery.net/v1/track/TRACK_ID/variant",
    headers=headers,
    json=payload,
)
data = response.json()

The track re-enters processing while Opus is built, then returns to ready. Parameter reference: Variants API.

3. Drop the original when you are happy

While you are still tuning recipes, keep the Original Upload so you can regenerate variants. Once voice sounds right and you no longer need re-processing, delete the original to reclaim the big WAV (or other source) from storage. After that, AudioDN cannot mint new variants for that track — only the remaining files (like voice) stay available for playback and download.

In the dashboard, open the track → files / variants, confirm voice is ready, then remove the Original Upload. See Original Upload for the trade-off: originals are insurance; after you are happy with the Opus result, they are mostly expensive nostalgia.

4. Play only the voice variant

Point the player at the track and request variants=“voice” with a Client-Side Player key. Listeners never download the multi-megabyte source — just the Opus file you decided to keep.

npm install @audiodn/components

<script type="module" src="https://unpkg.com/@audiodn/components@latest/dist/player.js"></script>

<script type="module">
  import '@audiodn/components/player';
</script>

<audiodn-player
  api-key="YOUR_PLAYER_API_KEY"
  scope="track"
  id="9b46405d-edc0-497c-9c36-4c97ceb15337"
  variants="voice"
  size="regular"
></audiodn-player>

Live demo

Playback is the voice Opus variant of the answering-machine sample — hear the footprint you get to ship.

More player options: Player component docs.

Checklist

• Upload the highest-quality voice source you have (WAV shown; any supported format works).

• A voice transcode exists with codec: “opus”, mono, and 32 (or 64) kbps.

• Listen to variants=“voice” and confirm speech quality is good enough for your product.
• Delete the Original Upload when you no longer need to regenerate variants — keep only the cheap Opus archive.